update readme

README.md

@@ -10,4 +10,120 @@ tags:
 - plant-disease
 size_categories:
 - 10K<n<100K
----
+---
+
+# Dataset Summary
+## Table of Contents
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
+  - [Languages](#languages)
+- [Dataset Structure](#dataset-structure)
+  - [Data Instances](#data-instances)
+  - [Data Fields](#data-fields)
+  - [Data Splits](#data-splits)
+- [Dataset Creation](#dataset-creation)
+  - [Curation Rationale](#curation-rationale)
+  - [Source Data](#source-data)
+  - [Annotations](#annotations)
+  - [Personal and Sensitive Information](#personal-and-sensitive-information)
+- [Considerations for Using the Data](#considerations-for-using-the-data)
+  - [Social Impact of Dataset](#social-impact-of-dataset)
+  - [Discussion of Biases](#discussion-of-biases)
+  - [Other Known Limitations](#other-known-limitations)
+- [Additional Information](#additional-information)
+  - [Dataset Curators](#dataset-curators)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+  - [Contributions](#contributions)
+
+## Dataset Description
+
+### Dataset Summary
+
+This dataset consist of 21,384 2448x2048 pairs of synthetic images for tomato plants. Each pair consist of left/right RGBD, and panoptic segmentation labels for the left image.
+
+### Supported Tasks and Leaderboards
+
+- `image-segmentation`: Both panoptic and semantic labels for separating tomato plants and identifying features and disease types in the dataset.
+- `depth-estimation`: ground truth depth values for stereo and monocular applications.
+
+### Languages
+English
+
+## Dataset Structure
+
+### Data Instances
+Each datapoint consist of 6 images:
+```
+{
+    'left_rgb': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=2448x2048 at 0x7F63FB5F4350>, 
+    'right_rgb': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=2448x2048 at 0x7F63FF2B3950>, 
+    'left_semantic': <PIL.PngImagePlugin.PngImageFile image mode=L size=2448x2048 at 0x7F63FC4488C0>, 
+    'left_instance': <PIL.TiffImagePlugin.TiffImageFile image mode=I;16 size=2448x2048 at 0x7F63FC497EF0>, 
+    'left_depth': <PIL.TiffImagePlugin.TiffImageFile image mode=F size=2448x2048 at 0x7F63FACF6E70>, 
+    'right_depth': <PIL.TiffImagePlugin.TiffImageFile image mode=F size=2448x2048 at 0x7F63FACF7560>
+}
+```
+
+### Data Fields
+- 'left_rgb': Left RGB image, was compressed to 95\% quality.
+- 'right_rgb': Right RGB image, was compressed to 95\% quality. Note the baseline is 3.88112 cm and HFOV is 95.452621 degrees.
+- 'left_semantic': Rendered colors that denotes the RGB label for individual pixels. See `example_load.py` for classes and sample scripts.
+- 'left_instance': Rendered colors that denotes the tomato plant instances for individual pixels.
+- 'left_depth': Rendered left depth compressed to 16-bit floats (in centimeters).
+- 'right_depth': Rendered right depth compressed to 16-bit floats (in centimeters).
+
+### Data Splits
+80/20 as shown in the train.txt and val.txt.
+
+## Dataset Creation
+
+### Curation Rationale
+Created to provide dataset for dense plant disease detection for a robotics platform with corresponding camera sensors and strobe lighting.
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+We used PlantVillage Dataset with further processing to align the healthy leaf colors with the purchased assets. We collected 750GB of original data where we compressed the depth images from 32-bit to 16-bit and RGB to 95\% quality for ~160GB.
+
+#### Who are the source language producers?
+See PlantVillage Datasets for tomato diseases. The tomato plants were purchased through SketchFab with modifications for extra green tomatoes and denser leaves.
+
+### Annotations
+#### Annotation process
+Annotations automatically generated through the simulation.
+
+#### Who are the annotators?
+Same as data creators with tomato leaf diseases the same as PlantVillage creators.
+
+### Personal and Sensitive Information
+[More Information Needed]
+
+## Considerations for Using the Data
+
+### Social Impact of Dataset
+
+[More Information Needed]
+
+### Discussion of Biases
+
+[More Information Needed]
+
+### Other Known Limitations
+
+[More Information Needed]
+
+## Additional Information
+
+### Dataset Curators
+[More Information Needed]
+
+### Licensing Information
+CC BY-NC-SA-4.0
+
+### Citation Information
+[More Information Needed]
+
+### Contributions
+[More Information Needed]

example_load.py

@@ -0,0 +1,88 @@
+from datasets import load_dataset
+import matplotlib.pyplot as plt
+import numpy as np
+from scipy import stats
+# similar to cityscapes for mmsegmentation
+# class name, (new_id, img_id)
+semantic_map = {
+                "bacterial_spot":       (0, 5),
+                "early_blight":         (1, 10),
+                "late_blight":          (2, 20),
+                "leaf_mold":            (3, 25),
+                "septoria_leaf_spot":   (4,30),
+                "spider_mites":         (5,35),
+                "target_spot":          (6,40),
+                "mosaic_virus":         (7,45),
+                "yellow_leaf_curl_virus":(8,50),
+                "healthy_leaf_pv":      (9, 15),  # plant village healthy leaf
+                "healthy_leaf_t":       (9, 255), # texture leaf (healthy)
+                "background":           (10, 0),
+                "tomato":               (11, 121),
+                "stem":                 (12, 111),
+                "wood_rod":             (13, 101),
+                "red_band":             (14, 140),
+                "yellow_flower":        (15, 131)
+                }
+
+def maj_vote(img,x,y,n=3):
+    half = n // 2
+    x_min, x_max = max(0, x - half), min(img.shape[1], x + half + 1)
+    y_min, y_max = max(0, y - half), min(img.shape[0], y + half + 1)
+    
+    window = img[y_min:y_max, x_min:x_max].flatten()
+    window = window[window != 255]
+    
+    if len(window) > 0:
+        # Perform majority voting
+        most_common_label = stats.mode(window, keepdims=True)[0][0]
+        return most_common_label
+    else:
+        return semantic_map["background"][0]
+    
+def color_to_id(img_semantic, top_k_disease = 10, semantic_map = semantic_map):
+    semantic_id_img = np.ones(img_semantic.shape) * 255
+    disease_counts = []
+    # remap rendered color to semantic id
+    for _, id_value_map in semantic_map.items():
+        # track disease pixel counts for top_k_disease filtering
+        if id_value_map[1] < 60 and id_value_map[1] > 1:        
+            disease_counts.append(np.sum(np.where(img_semantic == id_value_map[1], 1, 0)))
+        semantic_id_img[img_semantic == id_value_map[1]] = id_value_map[0]
+    # filter for most common disease labels
+    for i, item_i in enumerate(np.argsort(disease_counts)[::-1]):
+        if i >= top_k_disease:
+            id_value_map = list(semantic_map.items())[item_i][1]
+            semantic_id_img[img_semantic == id_value_map[1]] = 255
+            
+    # Apply majority voting for unlabeled pixels (needed as the rendering process can blend pixels)
+    unknown_mask = (semantic_id_img == 255)
+    for y,x in np.argwhere(unknown_mask):
+        semantic_id_img[y, x] = maj_vote(semantic_id_img, x, y, 3)
+    return semantic_id_img
+
+
+dataset = load_dataset("xingjianli/tomatotest", 'sample',trust_remote_code=True, num_proc=4)
+print(dataset["train"][0])
+
+
+left_rgb_img = dataset["train"][0]['left_rgb']
+right_rgb_img = dataset["train"][0]['right_rgb']
+left_semantic_img = np.asarray(dataset["train"][0]['left_semantic'])
+left_instance_img = np.asarray(dataset["train"][0]['left_instance'])
+left_depth_img = np.asarray(dataset["train"][0]['left_depth'])
+right_depth_img = np.asarray(dataset["train"][0]['right_depth'])
+plt.subplot(231)
+plt.imshow(left_rgb_img)
+plt.subplot(232)
+plt.imshow(right_rgb_img)
+plt.subplot(233)
+plt.imshow(color_to_id(left_semantic_img))
+plt.subplot(234)
+plt.imshow(np.where(left_depth_img>500,0,left_depth_img))
+plt.subplot(235)
+plt.imshow(np.where(right_depth_img>500,0,right_depth_img))
+plt.subplot(236)
+plt.imshow(left_instance_img)
+plt.show()
+
+